Agentic AI Engineer)

Socket.dev; New York (NY); On-site; USD 107,250 - 214,500; Full time

Benefits offered by this job

Health insurance

401(k) with company match

Paid time off

Job summary

Catapult is seeking an Agentic AI Engineer to design and build the specialist AI agents powering its platform, enabling athlete performance insights across sports.

You will develop a scalable workflow engine and a decision intelligence layer, ensuring calibrated, defensible recommendations with escalation where needed, all in production environments.

Qualifications

• 5+ years in applied ML or AI engineering with at least 2 years building production agentic AI systems
• Deep experience with multi-agent frameworks and orchestration
• Hands-on experience with confidence calibration and evaluation frameworks for probabilistic systems
• Production RAG experience with reranking
• Experience fine-tuning or adapting foundation models for specific domains
• Strong Python, Golang; experience with LLM observability and drift detection in production

Responsibilities

• Design and build the specialist AI agents that form the core of the platform’s intelligence layer — each one reasoning over a different dimension of athlete performance
• Develop the workflow engine that encodes domain scientist expertise into validated, versioned agent skills at scale
• Architect and build the decision intelligence layer that sits between agent outputs and practitioner delivery with escalation logic
• Enable multiple specialist agents to work together on complex questions, delivering traceable, calibrated recommendations within seconds

Skills

Python

Golang

Multi-agent frameworks

Confidence calibration

Production AI systems

LLM observability

Drift detection

AWS

GraphQL

REST

gRPC

Postgres

Mongo

Tools

AWS

GraphQL

REST

gRPC

Postgres

Mongo

Job description

Catapult is building the future of sports performance technology, with a mission to Unleash the Potential of every athlete and team on earth. We don\u2019t just work in the sporting industry; we are actively changing it. Since 2006, our solutions have been leading the way in sports performance software, science, and data, in a world where 1% can literally mean the difference between winning and losing.

We work with over 5,000+ teams around the world, empowering coaches, managers and trainers in premier teams in the NFL, NBA, NHL, MLS, EPL, AFL, NRL, NCAA and more. We provide the information they need to optimize athletes’ health, game-day readiness, and performance, as well as in-game tactics.

Catapult is a sports technology company that empowers professional teams to make data-driven decisions. We deliver health, performance, video, and AI insights from the locker room to competitive environments, ensuring every decision is an opportunity to gain an advantage, sharpen performance, and build lasting success.

WE WANT PEOPLE WHO ARE PASSIONATE ABOUT BUILDING AND SHIPPING AGENTIC SYSTEMS

The Agentic AI Engineer is a pivotal role in building the AI layer that compounds everything Catapult has ever measured. The goal is ambitious: to become the indispensable intelligence partner for every coach and athlete in every sport — fielding a bench of AI specialists that can each reason over a different dimension of performance and answer, together, the questions no single human analyst could assemble in real time.

This is the agent build role. You will design and build the specialist agents that form that bench, the workflow engine that encodes domain scientist expertise into validated agent skills at scale, and the decision intelligence layer that transforms agent outputs into calibrated, escalation-aware recommendations a practitioner can trust and act on.

Building an agent is not the hard part. Making an agent trustworthy — calibrated, grounded, escalation-aware, and provenance-traced — is the hard part. That is the standard this role is held to, and the reason it matters.

If you have shipped agentic systems in production — not demos, not prototypes — and you care deeply about what it means for a system to actually earn trust rather than assume it, this is the role where that experience compounds.

WHAT YOU WILL DO

• You will design and build the specialist AI agents that form the core of the platform’s intelligence layer — each one reasoning over a different dimension of athlete performance, from strength and conditioning to recovery, readiness, and beyond.
• You will develop the workflow engine that encodes domain scientist expertise into validated, versioned agent skills at scale, working directly with sport scientists to translate their judgment into calibration signals the system can act on reliably.
• You will architect and build the decision intelligence layer that sits between agent outputs and practitioner delivery — combining confidence weighting, consequence classification, and human escalation logic to ensure every recommendation is defensible before it reaches a coach or performance director.
• You will build toward the platform’s signature product experience: multiple specialist agents working together on a single complex question, synthesizing their findings into one coherent, traceable, calibrated recommendation in seconds.

WHAT YOU\u2019LL NEED

• 5+ years in applied ML or AI engineering, with at least 2 years building production agentic AI systems — not chatbots, not RAG pipelines alone, but systems with memory, tool use, multi-step reasoning, and calibrated outputs
• Deep experience with multi-agent frameworks and orchestration: dependency-aware routing, specialist agent composition, response synthesis across conflicting outputs
• Hands-on experience with confidence calibration and evaluation frameworks for probabilistic systems — you understand Platt scaling, isotonic regression, and ECE, and you have built evaluation harnesses that run against full input distributions
• Production RAG experience with reranking — you know that retrieval quality determines answer quality and have built pipelines that prove it
• Experience fine-tuning or adapting foundation models for specific domains — knowledge injection, not general text
• Strong Python, Golang; experience with LLM observability and drift detection in production

STRONGLY PREFERRED

• Experience building knowledge acquisition workflows for domain-specific AI — annotation interfaces, version-controlled knowledge bases, review queues, regression testing against skill updates
• Background working with domain scientists or clinical practitioners to encode expert knowledge into AI systems — you know how to translate judgment into calibration signals
• Experience with human-in-the-loop architectures: escalation models, confidence thresholds, consequence classification
• Familiarity with sports science, biomechanics, or performance data — understanding what "acute-to-chronic workload ratio" means matters in this role
• Experience with causal or counterfactual reasoning in AI systems — not just pattern matching
• Experience working with AWS (ECS, EC2, Lambda, SNS, SQS, etc), GraphQL, REST, gRPC, Postgres, Mongo

WHAT YOUR SUCCESS WILL LOOK LIKE

• Most agentic AI roles ask you to build systems that answer questions. This role asks you to build a system that knows when it does not know, and what to do about it. Human escalation is not a feature added at the end. It is a first-class architectural principle that runs through every recommendation the platform produces. The practitioner always owns the decision. Your job is to make that promise architecturally real.
• Equally important is the knowledge acquisition engine. Encoding domain scientist expertise into validated, versioned, production-ready agent skills, at the speed and scale that a multi-sport, multi-agent platform requires, is a genuinely hard engineering and process problem. The workflow you design determines how fast the bench grows and how trustworthy each new agent is from day one.

WHY CATAPULT?

• Catapult has spent twenty years collecting ground-truth athlete data from hardware on the body and on the field, across 40+ sports and 100+ countries. The domain scientist relationships that encode that data into agent skills have been built over the same period. The CEO has made the AI platform the central strategic bet for the next chapter of the company.
• The architecture is designed. The sequencing is clear. The team is being built now. What the platform needs is the engineer who can take proof-of-concept agents and make them production-ready, calibrated, defensible, and compounding in value with every recommendation made on the platform.
• Without calibration, intelligence becomes prediction. With calibration, intelligence becomes trust. That is what this role builds.

Compensation & Benefits

• The target Total Compensation range for this position is $107,250 - $214,500 per year. This range is inclusive of base salary and a target incentive plan (which may include equity, commission, or other bonus structures).
• Your specific compensation within this range will be determined by factors such as your geographic location, relevant experience, and job-related skills.
• In addition to this compensation, Catapult also offers generous paid leave and recognized company holidays, and the opportunity to participate in our comprehensive benefits package, including Health, Dental, and Vision insurance, and 401(k) retirement plan with company match.
• Whether you\u2019re interested in sports or not, you\u2019ll have the satisfaction of knowing your work is supporting some of the most successful teams and athletes on the planet!

All offers of employment are subject to Catapult's positive prehire check.